biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Prevent the cross-encoder logic from being applied to classification tasks (#18838)

Browse Source 
Signed-off-by: Max de Bayser <mbayser@br.ibm.com>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
This commit is contained in:
Maximilien de Bayser
2025-05-28 23:16:17 -03:00
committed by GitHub
parent 269d901734
commit 515b413ebf
5 changed files with 40 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/model_executor/models/roberta.py
View File

@@ -9,7 +9,7 @@ from torch import nn
from transformers import RobertaConfig
from vllm.config import VllmConfig
from vllm.model_executor.layers.pooler import CrossEncodingPooler
from vllm.model_executor.layers.pooler import ClassifierPooler
from vllm.model_executor.layers.vocab_parallel_embedding import (
VocabParallelEmbedding)
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
@@ -186,7 +186,9 @@ class RobertaForSequenceClassification(nn.Module, SupportsCrossEncoding,
embedding_class=RobertaEmbedding,
add_pooling_layer=False)
self.classifier = RobertaClassificationHead(config)
self._pooler = CrossEncodingPooler(config, self.classifier)
self._pooler = ClassifierPooler(vllm_config.model_config,
self.classifier)
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):
bert_weights, task_weights = roberta_task_weights_filter(weights)